$1B in transactions, 1 minute latency: dbt as production banking infrastructure
Keep's entire data team is one person, and dbt, Artie, and Snowflake together took its credit underwriting from a one-business-day, manual process to a sub-60-second decision.
Overview
Keep, a Canadian fintech serving 5,000+ businesses, uses dbt for analytics and real-time credit decisions, processing $1B+ in card volume with one-minute data refresh via Snowflake. Covers low-latency modeling, safe model design for critical workloads, cost/freshness/reliability tradeoffs, and audit trails via snapshots.
What you'll take away
- 01
A one-person data team moved credit underwriting from a one-business-day, spreadsheet-driven process to decisions delivered in under 60 seconds, around the clock, by buying infrastructure instead of building it.
- 02
Real-time CDC (Artie) and Snowflake dynamic tables handle streaming, while dbt handles model governance, versioning, and deduplication logic for customers who reconnect bank accounts across overlapping lookback windows.
- 03
Every underwriting calculation is stored alongside the git hash of the code that produced it, letting the team reprocess the entire portfolio against a new risk model with a single line of SQL.
Session speaker
Scott Ziolko
Keep